Would you use P-value to help make bets

In inferential statistics, we are taught the concept of a P-value, which is the probability of getting the result you got (or more extreme results) assuming the null hypothesis is true. In the casino industry, I was told by very knowledgeable people that they use an arbitrary P-value of 5% (which is the industry standard) to help discern whether the game is random or not, e.g. the casinos test for non-randomness and at a P-value of 5% or less, they deem the game to be non-random.

In a nutshell, they collect data, run a Chi-Square test goodness of fit, figure out the DF (degrees of freedom), and then get the P-value.

In the e-craps game by Aruze, the machine lists the last 15 rolls (unless one of the roll(s) didn't count then that roll or rolls would have a blank or dark space). I wrote a spreadsheet using Numbers on my iOS device to calculate the "historical" P-value of the last 15 rolls. The historical P-value DOES NOT predict the next roll.

Then I realized 15 rolls is a bit hard to update on the fly and make (hedging) bets within a 20-second window so I now use the last 12 rolls. Yes, I am calculating the Chi-Square test on the fly and it spits out the P-value. For example, I have found during a "hot shooter" where the machine has a red glow and the shooter(s) have made at least 20 rolls, the P-value goes down to the low teens or single digits.

Suppose the following 12 numbers came out were: 5, 7, 7, 7, 10, 6, 12, 3, 12, 7, 8, and 8.

I get a Chi-Square stat of 14.25 and P-value of 16.19%; my spreadsheet also tells me that there 1 roll of 4 or 10 short vs expectations, 1.67 rolls of 5 or 9 short vs expectations, 0.33 roll of 6 or 8 short vs expectations, 2 rolls of 7 over vs expectations, and 1 roll over for the remaining numbers (2, 3, 11, and 12) vs expectations.

I am using total dependent numbers when calculating the Chi-Square stat, and I got the DF as 10. Since there are only 11 ways to explain the outcome of two dice based on total dependent numbers: 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, and 12, then the DF is simply the number of explanatory variables less 1 or 11 minus 1 or 10 (the columns & rows approach should also get 10 for DF). Btw, there are 36 ways to explain the outcome of two dice using composition dependent numbers but you need to be able to discern the first die from the second, e.g. one die was blue and the second die was red -- but in the Aruze machine, it's impossible to discern the first die from the second, which is why I use the total dependent numbers approach.

What I do is when the P-value is 30% or less, I try to bet as little as possible (the irony is I don't make a lot of money during a hot shooter mode). When the P-value is 80% or more, I will go to an unhedged position based on the expectations. Between 30% & 80%, I try to make hedged bets. Based on hundreds of rolls that I have tracked, the "rolling" P-value goes in cycle, it doesn't stay in a specific range like P-value of 80% for the next 20 rolls. That "rolling" P-value bounces a lot and can change directions on a dime.

Basically, I am taking a position only when the P-value is 80% or higher, e,g, betting that the expectations will continue.

Is this something you would use to help make bets?

you seems to know your stat 101 well and you already said "The historical P-value DOES NOT predict the next roll."

I don't know where you are heading

My understanding of using hypothesis methods is trying to determine whether something is as expected (in your case random) or something is up. You may also come across Bayes theorem but that is a different idea.

Yes you have to set a confidence level but you also have to run lots of trials if you want some confidence in your findings. For instance if you were trying to prove an on-line site was dodgy, the dice were biassed, the roulette wheel wasn't level you'd need to really make sure. In this case 95% isn't enough.

Simple scenario - a factory is making 6" pencils and has a obligation to ensure they're over 6" long, but their actual length is random. An inspector wants to check whether they're meeting standards so will take a batch of N pencils and measure them to see whether they're selling short. This is done by looking at the average length and standard deviation of the sample to estimate the chances the pencils are short or your sample was just bad luck.

Remember this method requires you to state your test before you take the sample. Naturally if you conduct a test and prove the dice are biassed then you can adjust your expectation and bet accordingly - however in practice casinos keep changing the dice.

Jason



Long time ago Ahigh asked me for some data to analysis. I cant remember the
number of rolls I sent him. It was a decent number.

When I looked at the data, it was fairly typical I thought with some indication
of a outside bias. Ahigh came back with some crap that this data was way to
good to be true. no one could do that, the P values were off the charts.

Well I did not even know what a P value was. I then sent the same data to
a real good craps guy and he ran the numbers and felt just like I did. these
were decent numbers, with a small out side bias. Nothing you could go to the
table with and make a ton of money.

So I still don't know what a p value is, however if it is supposed to be a guide
for betting, and it was applied to my data correctly, betting only on that is a sure
way to get to the poor house.

dicesetter

DiceSetter,
Seems you had a good time (maybe not the best of times). You could, you know, take some time out from your practice roll and write-up a trip report. I would love to read it, I suspect I am not alone. You can save the best of the table stuff for other posts. Where'd you go? You run into any suspected WoV members? 2F

I did, it is the charting thread

dicesetter

Quote: dicesitter

Well I did not even know what a P value was. I then sent the same data to
a real good craps guy and he ran the numbers and felt just like I did. these
were decent numbers, with a small out side bias. Nothing you could go to the
table with and make a ton of money.

Nothing you could go to the
table with and make a ton of money.

As in YOU ALL non AP DI's.(each year the DI community gets smaller due to the facts and disappearing BR's)

As ME has pointed out many times all you supposed DI's have no clue how to make the proper bets in order to maximize your supposed "skill" .

After 15 years if someone had a small advantage it would've blown up and turned into millions and surfaced by now. Explain why the casinos are now embracing DI?

If for some reason the casinos closest to you or whatever places you frequent are less tolerant. Spend 6 mo to 1 year in Vegas and make millions.

24/7 365 multiple location advantage without heat. WTF you guys waiting for? That's an AP dream.

Guys make millions on small BJ advantages.

Craps would be far better. Why isn't anyone making even a living playing craps?

Axel


Axel then you should go play blackjack.... surely a guy like you that
knows everything about everything should find it easy to get a edge
at BJ and my god if a 70 year old guy can affect the HA on a craps
table, you should find this a piece of cake.

But, it takes years of work to get an advantage on BJ and years to
get some advantage on craps..... a person can learn to talk with little
or no work.

dicesetter

Quote: dicesitter

...But, it takes years of work to get an advantage on BJ...

100% false.

Couple of things.......

First as Axel says if it is so easy where is your million??


Second Axel still cant figure the basic thing in regards to all of this
stuff.... The fight is not with the casino, It is with yourself, any one can
learn blackjack enough to get the basic idea of card counting, equally
easy is learning some type of repeatable craps shot.

Do you have enough control over yourself to put in the time to be
excellent at it, to control your betting, control your casino activity,
control your temper when they kick you out for card counting,
when they constantly tell you to shoot harder, shot higher
or lower, hit the back wall harder, put more air under the dice and
on and on.

Ibeatyouraces... I am sorry Axel can't learn card counting well enough
and cant learn a repeatable shot on craps, I really am, but if you cant
do that, how in the hell are going to control yourself when you have to.

dicesetter

My point was, you don't need to count to get an edge. Counting only determines when you have it. If we're both playing and I'm counting and you aren't, and at some point I recognize this and raise my bet, you ALSO have an edge too. You just don't know that you do.

Ibeatyouraces


I agree with that. But if the count is correct, your hand has
no better access to the cards than the dealers.

There are many things you can do in blackjack and
craps to better your odds. It is really not a lot different than
playing good poker.. it is on you.

I know a few of the better blackjack players, and they to a man
have suggested the rule changes, shuffle machines etc have made
it much harder to win. Even a double deck game where the
cut is almost a full deck hurts.

Ibeat, my point is that if you look at gambling and gamblers, the hardest
part is the control you have over yourself, your efforts and your
emotions. Why are casino's so afraid of card counters and dice setters,
it is because if you took the time to be good at that portion, you more
than likely have better control over your play. I think it is silly myself, 99%
of gamblers have control over nothing.

That's my problem, it is not throwing the dice, it is all the other stuff.

dicesetter

Quote: Jason14x

In inferential statistics, we are taught the concept of a P-value, which is the probability of getting the result you got (or more extreme results) assuming the null hypothesis is true. In the casino industry, I was told by very knowledgeable people that they use an arbitrary P-value of 5% (which is the industry standard) to help discern whether the game is random or not, e.g. the casinos test for non-randomness and at a P-value of 5% or less, they deem the game to be non-random.

In a nutshell, they collect data, run a Chi-Square test goodness of fit, figure out the DF (degrees of freedom), and then get the P-value.

...
Is this something you would use to help make bets?

You're using the correct mathematical model, but your logic is flawed. The only thing that could cause your expectation to change in any craps game would be a problem with the dice (assuming random rolls, and while these machines aren't perfect, they are probably fairly random rolls). You wouldn't try to detect a problem with the dice on a rolling basis - you would take a number (at least a few hundred) rolls and see what your P value is. If there is a problem with either die, the P value would continuously trend downward. There is a nice advantage to these machines in that it knows which die is which. The value reported in the roll history for the die shown on top is always for the same die, and vice versa. This could help you track each die independently and thus requires fewer rolls to come up with an accurate P-value for both dice. If one die is skewed, you can then run the math to determine what that means for the potential outcomes. It doesn't take that many rolls; the Wizard himself was able to determine absurdly high bias at an online craps site after just 328 bets.

I do have a hypothesis that the longer these large dice keep getting batted against plexiglass, one or both of the dice may in fact develop a bias. I have seen extremely statistically improbable runs on these machines. They aren't like regular craps tables; changing the dice requires a fair amount of work and is usually done less than once per week. You can tell if they are aged by simply looking at them; the new ones are very shiny, and after a few days they are grimy and have scratches all over them.

All of that said, I have run a chi-squared test on 2 machines at varying times looking at the results of each die independently and haven't come up with anything that would suggest outside bias one way or another. I tend to get P-values in the 50-60% range, and you'd need one with a P value of around 1/10th of that to determine bias with any accuracy. That doesn't mean it's not possible to find one with some old dice that have been chipped, or dice that came slightly imbalanced from the factory etc., that have a bias, but I haven't found any yet.

Finally, one word of caution. Having listened to Bob Nercessian (sp?) speak several times about this subject, I wouldn't be using your phone to track these things and make betting decisions in realtime. I have done it with pen and paper, the reason being that you may very well be running afoul of the device law in Nevada, in the same way that it is illegal to use a device to track and determine statistically perfect blackjack strategy plays. It probably wouldn't become an issue until you actually found a bias and beat them up with it, but you may be committing a felony by using your phone for this.